FR2886791A1 - Timing information recovering method for clock synchronization between master and slave nodes involves adjusting frequency of slave clock to match master clock using offset of time grid - Google Patents

Abstract

A series of timing packets is exchanged between the master node and slave nodes. An offset of time grid relative to the master and slave clocks is measured. The frequency of slave clock is adjusted to match the master clock using the measured offset of time grid. The difference between the time grids and the frequency offset is used to control a frequency synthesizer generating slave clock. An independent claim is also included for arrangement for recovery timing information.

Description

METHOD AND DEVICE FOR RECOVERING THE

  SYNCHRONIZATION ON A GRANULAR NETWORK OF

PACKET SWITCHING

  The present invention relates to a method and a device for retrieving synchronization information. It applies in particular in the field of packet switching networks.

  When temporary data, for example voice or video data, is conveyed between a source node (master) and a slave destination node on a packet switching network, the local clocks of the source node and the destination node are not synchronized. It is therefore necessary to use any method for recovering the source clock at the destination node, in order to retrieve the temporary data.

  The current clock or timing distribution systems, such as the IEEE1588 Network Time Protocol (NTP), which use packet switching networks to synchronize clocks, send packets called "synchronization packets" in both directions between the source node and the destination node. From these packets, the transmission time is compared to the arrival time to accurately determine the transit time of the packet. At the destination node, the clock frequency is adjusted to maintain a constant transit time measured for all synchronization packets passing through the network. In this way, it is possible to synchronize the clock at the slave units on the clock at the master unit. This process is called a Constant Transit Time (CTT) clock distribution.

  A method based on the CTT principle can combine the transit time for packets flowing in both directions between the source node and the destination node. Assuming that the paths are both subject to the same delay, it is possible to eliminate the constant transit time so that the slave clock operates without any offset from the master node. This technique is known by the name of "synchronization simultaneity" (Same-Time in the Anglo-Saxon literature).

  Current clock synchronization methods, such as the Network Time Protocol (NTP) or IEEE-1588 Precision Time Protocol (PTP) , are based on the fact that the packets circulate on the network with a constant transit time (CTT process). Such methods are sensitive to packet timing variations.

  The accuracy of the CTT process deteriorates when the synchronization packets are queued at the switching components and when the switching components have a high synchronization granularity, which means that the packets are dispatched at intervals of discrete times. On Ethernet networks, a granularity has been measured which correlates with the time required to send 3, 64 or 72 bytes over a 100 Mbps link. Networks based on the time division multiplexing (TDM) technique, for example Ti and E1, have a time granularity that is correlated with the 8 kHz frame rate.

  It is not possible to determine the transit time to a degree of accuracy that is greater than the associated timing granularity, and therefore the network granularity results in a loss of accuracy in the timing information that is exchanged between the source node and the destination node.

  The present invention overcomes the loss of precision that occurs when a temporal granular switching component, or a synchronous series of such components, imposes a limit on accuracy.

  The present invention provides a method for recovering synchronization information between a master node and a slave node, while the respective mother and slave clocks are interconnected on a packet switching network having an associated time grid with a distinct granularity, comprising the steps of exchanging a series of synchronization packets between said master and slave nodes or sending synchronization packets from a network component to said master and slave nodes; measuring an offset of said time grid with respect to the mother and slave clocks; and using said offset to create a clock at the slave node that matches with said master clock.

  According to one aspect of the invention, a distributed system is used to synchronize the frequency of the slave clocks to a master clock on a switched / routed packet-switched network. The system sends packets in both directions between the master and slave nodes connected to the packet switching network. At the moment of their arrival, the packets are subjected to a time stamp which is carried out by the local clock of the node which receives the packet. The packet-switched network contains a switch or router that routes these packets in distinct multiples of an associated time grid. The type of granular time grid corresponds to the type found in some existing switching and routing components.

  In one embodiment, it is possible to determine the granularity at the nodes that receive the packets by searching for a denominator value which provides a common remainder for a series of timestamps entered at the time of their arrival. A frequency offset of the local slave clock has the effect of creating an offset for the two values of the denominator found at the master station and the slave station. The system uses the offset found to adjust the slave clock frequency or to generate a new time base from the slave clock using frequency synthesis techniques. The system can be used in combination with existing time synchronization techniques to improve accuracy.

  This invention uses the network time grid as a common reference, and refers to the Reference Granularity (RG) clock distribution. It is possible to use the RG method autonomously to obtain a precise clock frequency distribution method, or it can be used in combination with a CTT method to achieve simultaneous accuracy.

  The measurement of the granularity of the time grid is the subject of discussion, for example, in the document "Modifications of the Euclidean Algorithm for Isolating Periodicities from a Sparse Set of Noisy Measurements" BMSadler, SD Casey (http: // www. : //techreports.isr.umd.edu / rep orts / 1995 / TR 95-105.pdf) and "On Periodic Pulse Interval Analysis with Outliers and Missing Observations" BM Sadler, SD Casey (http: // techre ports.isr .umd.edu / reports / 1996 / TR 96-6.pdf). These references show that taking only 10-100 time stamp samples for a short period of time is sufficient to determine the granularity that is present in the measurements. In the presence of noise, more samples may have to be taken, although there is a deterioration in performance if there are too many samples.

  The method based on the reference granularity (RG) object of the invention is superior to the CTT method when temporal granular components are part of the data path taken by the synchronization packets. A CTT process would need a very large number of timing packets to spread the average of the noise that is embedded in the transit time measurements. Any calculation of the average requires an interval of time to acquire precision. The higher the degree of precision required, the greater the number of samples that are required, and based on a constant sampling rate, the longer the required time interval will be. There is, however, a limit to the amount of time that can be devoted to averaging in order to enhance accuracy, since during this time the slave clock could begin to suffer a drift effect and thus risk introducing another source of inaccuracy.

  A drift effect will also deteriorate the sensitivity of the RG process. However, experiments have shown that accuracy improves much more rapidly and reaches a level greater than 1 ppb (10-9) if packet arrival times of only 10-100 packets are used within an interval of less than 1 ppb (10-9). not exceeding a few seconds, and inaccuracy in the measurement of timestamps of more than 50 ns.

  Considered from the material design perspective, the approach adopted in the modified Euclidean algorithm, as stated in the aforementioned references, can be implemented by a person skilled in the art, without having recourse to a digital division operator or an operator 2886791 5 of module. The dataset needs consecutive iterations in terms of differentiation and sorting by segmentation. Their mapping can be done through the use of some registers, a subtractor, a comparator, a memory unit and a simple controller. If necessary, the mapping of these iterations can also be performed on a programmable processor.

  This invention also applies to an embodiment in which packets are transmitted from a third node to master and slave nodes, provided that these packets pass through the same time granular reference switching component. / routing. The third node does not necessarily participate in the synchronization process between the master and slave nodes, since it does not need information from these nodes.

  This invention also applies to situations in which the actual switching component sends time granular packets to the master and slave nodes. Alternatively, these packets may be transmitted in unicast, multicast, or broadcast mode, again provided that the packets are routed through a single common temporal granular reference component.

  The invention will now be described in more detail, by way of example only, with reference to the accompanying diagrams, in which: FIG. 1 represents a master node and a slave node that are connected to a communication network; Fig. 2 shows a timing chart showing the grid of time granularity of the network.

  Fig. 3 is a block diagram of an embodiment of the apparatus, according to the present invention, for measuring the time granularity grid by involving a series of time-stamps without noise; Figure 4 is a block diagram illustrating an apparatus, in accordance with an embodiment of the present invention, for reconstructing a master clock incorporating a frequency synthesizer; Fig. 5 is a block diagram illustrating an apparatus, according to an embodiment of the present invention, for recovering the mother clock by adjusting the frequency of the slave clock.

  Fig. 6 is a block diagram which illustrates an apparatus, in accordance with an embodiment of the present invention, for determining an approximate estimate of the temporal granularity grid that is present in a given set of noise-loaded timestamps; Fig. 7 is a block diagram illustrating an apparatus, in accordance with an embodiment of the present invention, for implementing a simplified method for determining a complex exponential term; and Fig. 8 shows the result from measurements of arrival times that have been made from packets flowing through a granular timing switch component.

  This invention describes a method for synchronizing clocks between master and slave nodes when changes in transit time of a packet are a cause of a time-granular switching and routing component, which packet order for their output according to distinct multiples of an associated temporal grid. The method described also applies when a series of such components operate on a single common time grid, as in the SDH and TDM networks.

  A group of nodes is connected to a communication network. The network contains a switching / routing component that routes these packets in multiples that are distinct from an associated time grid. The network could also contain a series of these components that are responsible for routing packets based on a common time grid. Each node is equipped with a local clock which serves to determine the arrival time of a packet sent from the network. Slave nodes send packets to the master node over the network. The master node transmits the arrival time of the packet to the original slave node. The slave node measures the arrival time of the packet in question.

  Synchronization packets should be sent at fine-grained random times to prevent the formation of temporal granularity that is introduced by the actual invented clock distribution system.

  Each slave node measures the granularity of the arrival time and compares it to the granularity of the arrival time at the master node. It is the offset between the two values that will determine the frequency offset of the slave clock. It is possible to calculate the frequency of the slave clock from the following equation: FS = FM * GM / GS, where Fs is the frequency of the slave clock, FM the frequency of the master clock, GM la granularity at the master and Gs the granularity measured at the slave node, the frequencies are in Hertz and the granularities in seconds.

  The reference granularity (RG) is based on the analysis of the arrival times of packets on packets arriving at both the master node and the slave node, which were routed through a single common component switching / routing with temporal granularity. The associated time grid is determined from these arrival times.

  The temporal granularity grid can be determined from noise-free timestamp samples, using an initial time stamp to, a first time stamp ti, and a second time stamp t2, which allows the calculation of the Greater Common Divider (GCD). ) of (t1to) and (t2-to) for example using the device of Figure 3. The time grid measured at the master station and the slave station will have a slight offset which is linked, on the linear plane, to offset of the frequency between the mother clock and the slave clock. The offset of the time grid thus measured is then used to generate a clock devoid of any frequency offset at the slave station.

  2886791 8 The process is robust for noise measurement, and provides a precise result. Experiments have shown that accuracy greater than 1 ppb (10-9) can be achieved if the arrival times of 10-100 packets are used within a few seconds.

  Referring now to FIG. 1: a master node 1 and a slave node 2 are each connected to a communication network 3 which includes a time-granular switching / routing component 4, or a series of such components. The master node 1 and the slave node 2 exchange packets via the network 3. The component 4 works in conjunction with a time grid and thereby introduces granularity into the network 3.

  Figure 2 shows a timing chart that illustrates the time granularity grid 31, as measured with respect to the master clock (tM) and the slave clock (ts). The intervals gM and gs represent the granularity intervals that have been measured with respect to the mother and slave clocks respectively. The arrival of the packets on this grid, as measured with respect to the respective mother and slave clocks, is illustrated by the vertical arrows 32.

  In order to measure the temporal granularity grid, a series of packets is sent in both directions via the network 3 between the master and slave nodes. On arrival, packets receive a time stamp from the local clock associated with the node receiving the packets. The packets arrive at the receiving node at times which are multiples distinct from the granularity intervals of the time grid of the network, the latter having a distinct granularity, as noted previously. The receiving node then attempts to find a denominator value that provides a common remainder for the series of timestamps entered upon arrival. As can be seen in FIG. 3, the timestamps are conveyed by delay lines 5 and summers 6 before being injected into a circuit 7, which determines the Largest Common Divider (GCDC) of the series of the arriving packets. .

  Once the common denominator has been found, the existing frequency offset on the local slave clock will cause an offset on the two denominator values found at the master station and the slave station. The system employs the offset having been found to adjust the slave clock frequency or to generate a new time base from the slave clock using frequency synthesis techniques. The system can be used in combination with existing time synchronization techniques to improve accuracy.

  Fig. 4 is a block diagram illustrating an apparatus according to the present invention for reconstructing a master clock. This apparatus uses a frequency synthesizer 11 which multiplies the frequency of the slave clock 12 by the GS / GM ratio of the number representing the slave time grid 9b to the number representing the time grid of the master station 9a. This value is produced by a divider 10, which receives the inputs 10a, 10b from the respective time gate units 9a, 9b. The slave time grid is obtained by routing the incoming timestamps to the timestamp unit 8 which time stamps the incoming packets according to the slave clock. Frequency synthesizer 11 is responsible for providing a reconstructed master clock which can be seen by using a frequency synthesis technique which multiplies the slave clock with the GS / GM ratio.

  Fig. 5 shows a block diagram which illustrates an alternative embodiment where the outputs of the time grid units 9a, 9b are subtracted in a subtractor 6 to obtain a difference signal which is fed into a circuit control unit This action makes it possible to drive a digitally controlled oscillator 14 to produce the recovered master clock which provides the local clock to the slave node. The loop controller 13 reduces to zero the difference 6 (GM Gs) between the mother time grid having been measured 9a and the slave time grid having been measured 9b.

  Figures 4 and 5 show two examples of a possible implementation of the invention described herein. The example in Figure 5 can be mapped to the hardware. The example of Figure 4 requires division, whose execution can be done on a programmable processor.

  Figure 6 is a block diagram illustrating an apparatus for determining an approximate estimate of the temporal granularity grid that is present in a given set of noise loaded timestamps. The incoming timestamps are routed through a multiplexer 15, and the difference between the transit time of a current packet and a timed packet is in the subtractor 6. Packets that are below a certain threshold are eliminated by a unit 16, while the remaining packets are sorted in a unit 17.

  It is also possible to obtain the grid of the temporal granularity in the presence of instability effects and samples of noise-laden time stamps, thanks to an iterative process carried out on a dataset which starts with a series of timestamp samples sorted by value (sort) and which operates by performing the following steps of: a) updating the dataset by differentiating the samples in the dataset; b) eliminate all samples below a given threshold level. If only one sample remains, this will be the estimate for the temporal granularity grid. Then we do the iterations.

  c) sorting the given set of samples according to the value; d) add a zero sample to the sample set and maintain the game in sorted order; e) continue the repetition of the iteration indicated in step a) above.

  The threshold level is an optimization factor: it should be smaller than the expected minimum granularity level, and it should be larger than the amount of noise expected in the measurements. The second and third steps b and c above may be switched, but this may be less effective.

  An accurate estimate of the temporal granularity grid can be obtained if we use an initial estimate for "G", when we run an iterative process on a dataset that starts with a series of sorted timestamp samples. by value (t) and operating on the following basis: a) updating the dataset by differentiating the samples in the dataset; (b) remove all samples below a given threshold level until only one sample remains; c) sort the samples according to the value; d) add a zero sample to the sample set while maintaining the game in sorted order; e) continue the repetition of the iteration indicated in step a. It is then possible to further refine the value for the granularity "G" by virtue of the optimization of the given series "A" of time stamps in the complex exponential equation given by: 2ErEiEG A fast process, used to refine the value of G, consists in calculating two results A1 and A2 from the equation above for two neighboring values G1 and G2, and provided that G1 <G <G2 is verified. The refined value for G is then obtained on the basis of the equation given below: G = A; .G, + Az.G., A, '2 + A; A simplified method for calculating the sum of the complex exponential term in the formula above is to use timestamps having integer values, a division, and a look-up table that contains a sampled and scaled version relative to a complex unitary circle representative of the exponential term (Figure 7). For each timestamp, we take a pair of real numbers Re imaginary, lm from a lookup table 20 which contains 16 pairs of numbers of this kind. The table is addressed with the 4-bit integer remainder from division 10 of time stamp 18 multiplied by 16 by the value of G. The series of numbers Re is summed and then squared; the series of numbers lm is added and then squared. Finally, we add up the results that were squared to obtain the squared amplitude (A2) of all the Re, lm vectors that were summed.

  If desired, it is possible to use an alternative look-up table capable of accepting any number of entries smaller or larger, or capable of taking further resolutions of samples Re, lm.

  The RG method, involving the above equation, loses in sensitivity when the drift of the clock frequency with respect to the measurement interval causes a time deviation that is close to the temporal granularity grid or is more important than the latter. The iterative process is less sensitive to drift of the clock but more sensitive to noise, which then results in a lack of precision. The greater the number of iterations required to produce a result, the greater the sensitivity to noise.

  Figure 8 shows the result obtained with actual measurements of the arrival times that were found on packets flowing through a granular timing switch component. Arrival times were provided by a module. These data can only be produced using an extremely accurate estimate of the granularity grid. Even a small deviation from this estimate can result in a thick band of measurement points that span the entire range of the applied module. The irregularities of the curve are one of the causes of the frequency drift affecting the clock of the packet switching component system.

  As can be seen, the invention makes it possible to recover a clock frequency at a destination node (slave), which is an exact copy of a source node (master), by sending packets on a packet switching network. A source node transmits packets to destination nodes while the destination nodes send packets to the source node. In the case of the packet switching network, it can be of any type, for example Ethernet, ATM or synchronous networks. This invention finds application in a network that contains a network routing / switching component designed to re-route packets at multiple distinct intervals of an associated time grid. This invention also finds its application in a synchronous network, in which case series of these routing / switching components re-route data in distinct and common multiples of an underlying time grid, for example SDH networks.

Claims (2)

13 Claims   A method for retrieving synchronization information between a master node (1) and a slave node (2), while the respective mother and slave clocks are interconnected on a packet switching network having a time grid (31) associated therewith a distinct granularity (gr and gs), comprising the steps of: - exchanging a series of synchronization packets between said master and slave nodes or sending synchronization packets from a network component to said master and slave nodes ; - measuring an offset of said time grid (31) relative to the clocks (32) mother and slave; and using said offset to create a clock at the slave node that matches with said master clock.   2. Method according to claim 1, characterized in that said offset is used to adjust the frequency of the slave clock so that it matches the master clock.   3. Method according to claim 2, characterized in that the difference between the time grids measured with respect to the mother and slave clocks is used to control the frequency of said slave clock.   4. Method according to claim 1, characterized in that said frequency offset is used to drive a frequency synthesizer which generates said slave clock.   5. Method according to claim 4, characterized in that the ratio of the ratio of the time gates measured with respect to the mother and slave clocks is used to control said frequency synthesizer.   Method according to claim 1, characterized in that the time gates are measured with respect to said mother and slave clocks, following the examination of the series of synchronization packets received at said respective master and slave nodes in order to find a denominator value that provides a common remainder for the series of packets received.   The method of claim 1, characterized in that the slave node sends said series of synchronization packets to the master node, while the master node transmits the arrival time of the packet, relative to the master clock. , back to the original slave node.   8. Method according to claim 1, characterized in that the synchronization packets are sent at fine-grained random times.   9. Method according to claim 1, characterized in that the frequency of the slave clock is calculated from the following equation: Fs = FM * GM./Gs, where FS is the frequency of the slave clock, FM the frequency of the master clock, GM the granularity at the master level and Gs the granularity measured at the slave node, the frequencies are in Hertz and the granularities in seconds.   10. Method according to claim 1, characterized in that it is used in the presence of instability effects and time-stamped samples loaded with noise, it comprises the execution of an iterative process on a dataset which starts with a series of timestamp samples sorted by value (tn) and operates on the following basis: a) update the dataset by differentiating the samples in the dataset; (b) remove all samples below a given threshold level until only one sample remains; c) sort the samples according to the value; d) add a zero sample to the sample set while maintaining the game in sorted order; e) continue the repetition of the iteration indicated in step a.   11. Method according to claim 1, characterized in that the grid of the temporal granularity for each node is obtained by means of an initial estimate G of the time grid according to the method set forth in claim 10 and by a further refinement of G by virtue of the optimization of the given series A of timestamps "t" in the complex exponential equation given by: rn ec 2886791 12. The method according to claim 11, characterized in that it seeks for A two results Al and A2 for two neighboring values G1 and G2, with the proviso that G1 <G <G2, and obtaining for G the refined value on the basis of the equation given below: A; A method according to claim 11, characterized in that the sum of the complex exponential term in the equation (2 r G) e is calculated by using timestamps having integer values, a division and a look-up table (20) which contains a sampled and scaled version of a complex unitary circle of the exponential term.   Apparatus for recovering synchronization information between a master node (1) and a slave node (2), while the respective mother and slave clocks are interconnected on a packet switching network having a time grid (31) associated with it a distinct granularity, comprising: - means (9a, 9b) for measuring an offset of said time grid (31) with respect to the mother and slave clocks (32); and and means (11) for creating a clock at the slave node which matches with said master clock, using said measured offset.   15. Device according to claim 14, characterized in that said slave clock comprises a numerically controlled oscillator which is controlled by said means to measure an offset of said time grid.   16. Device according to claim 15, characterized in that it further comprises a difference circuit (6) for generating a difference signal which drives said digitally controlled oscillator, said difference circuit receiving inputs respectively from said means (9a). for measuring an offset of said time grid with respect to the master clock and said means (9b) measuring an offset of said time grid with respect to the master clock.   17. Device according to claim 14, characterized in that it further comprises a frequency synthesizer (11) for creating a clock reconstructed from said slave clock and the ratio of time grids measured with respect to mother and slave clocks. .   Apparatus according to claim 14, characterized in that said means for measuring the time grid comprises a calculator (7) for determining the greatest common divisor of said series of synchronization packets which provides a common remainder.   19. Device according to claim 14, characterized in that said means for measuring the time grid comprise first and second delay units (5) connected in series to receive incoming packets, a first element to find the difference between an input of the first delay unit and an output of the second delay unit, and supply the result to a first input of the calculator (7) of the largest common divider, and a second element to find the difference between an input of the second delay unit and the output of the second delay unit, and outputting the result to a second input of said calculator (7) of the largest common divider.   20. Device according to claim 14, characterized in that said means for measuring the time grid comprise a look-up table (20) for producing real and imaginary parts of a complex number A for a series of timestamps tM where e G